Machine Learning Q&A

What are you using data for?

  • Our goal is to provide you with the best real-time sales recommendations on 1) which companies to target, 2) which prospects to reach out to, 3) when & how to target them & what to write.
  • Each company has different needs. While it is possible to create generic recommendations and messages, best results can only be achieved by personalizing the results based on your prospects, Closed Won Opp Contacts, Accounts and Customer Accounts.
    • That way, UserGems can generate custom models and custom insights that are always relevant and up-to-date.
  • What’s important to note here is that the training of the model is not in order to train a general model (which would be kept and e.g. used for other customers too). We already have a general model.The training of the model with your date is in order to train a custom model that is only used to provide services to you (and not in order to provide services to other customers).
  • We mainly need your data, so that we can create a custom model for your specific use cases - otherwise we would not be able to provide tailored suggestions (e.g. "you should focus on this persona") or, in case you want to benefit from our AI messaging, provide messages that are oriented towards your most successful messages.

What data from our CRM are you using?

From Salesforce, we use:

  • Contact information
  • Opportunity Contacts
  • Account information

How do you ensure that the “Model Data” does not identify Customer or Customer Data?

  • “Model Data” (the data representing the stored models), does not contain any sensitive data or any information about specific prospects or accounts.
  • Our models are trained on training data that is stripped of any sensitive or personal information. Any person names, email addresses or job-titles, as well as any company names, domains, descriptions, etc. are removed from our datasets. Instead, we use anonymized, classified data, such as job-title- or company categories.

What features can I not use if I do not agree to using my data to train Machine Learning ?

Here are some examples of features you cannot use:

  • Persona creation and improvement suggestions: Our machine learning model can automatically create the best Persona for your company, based on current Open Opp Contacts (= the prospects that showed interest). If you already have a persona, our machine learning model can suggest improvements.
  • ICP creation and improvement suggestions: Our machine learning model can automatically create the best ICP for your company, based on your Customer Accounts. If you already have an ICP, then our machine learning model can suggest improvements.
  • Ranking prospects based on conversion rate of similar Opportunity Contacts: Usually we find thousands of possible prospects for our customers. In order to decide which prospects to focus on, we create a company specific AI model, that ranks prospects based on the “likelihood to convert” (by comparing prospects with the Opportunity Contacts). If we cannot train our model with your specific data, we will not be able to rank prospects.
  • Email message: For each signal & prospect, we do suggest personalized text snippets. Based on how these snippets are used and how successful they are, your machine learning model will suggest different wording.
  • Signal suggestions: Based on the Contact and Opportunity data in your Salesforce, we can suggest which Signal will bring the most value. 

Please note, that we are continuously adding new Machine Learning features, so if your legal team rejects our usage of company data to train our model, it means that you will not have access to any existing or new machine learning developments. 

It is recommended that the legal team consults with the economic buyer before pro-actively removing Model Data from the Terms of Services, since it will significantly reduce the functionality UserGems is able to offer.

How do you ensure that no Personally Identifiable Information (PII) I upload to UserGems will be used in the prompt?

We do not use any PII you provided in our prompt and therefore do not send any PII to OpenAI. Names, email addresses or other PII you uploaded to UserGems will never be shared with OpenAI.

If names of people or companies are used in the prompts, they will be replaced with placeholders before sending the prompt to Open AI, and are then, after receiving the response from Open AI, reverted back on our servers.

What Information is used in the prompt?

Information directly used in the prompt:

  • Your company's value proposition (as provided by you)
  • Your buyer personas, with their specific pain points and your value proposition for them (as provided by you)

Information indirectly used in the prompt:

Signal information is used to adjust our prompt to specific use cases.

Examples:

  • If a prospect recently changed their job, we’d tell the prompt to start the message with “Congrats on the new gig, <placeholder_first_name>!”
  • If the target company is a competitor of one of your customers, we’d adjust the prompt to use “We've done this for <placeholder_customer_competitor_name> so thought it could work for you too.”

Signals that are currently used:

  • Past relationship with the prospect (e.g. past champion, past open opportunity contact)
  • Prospect-specific signals (e.g. recent promotion, recent job change)
  • Account-specific signals (e.g. the company is a competitor of one of your customers, the company recently received funding)

How is the message generated?

The message prompts are then used to generate sales messages tailored to your specifications. This is done in two steps:

Generate message candidates. Depending on the level of prompt customization, we do this by either

  1. Using our custom-trained models (which are trained without PII)
  2. Using generic OpenAI models

In either case, we use the OpenAI infrastructure.

  • Evaluate the messages candidates to choose the best message. This is done using a custom-trained model, again using the OpenAI infrastructure.

For both steps, the OpenAI security policies apply (see https://openai.com/enterprise-privacy/)

Was this article helpful?